Function formatFloat

src/compute-engine/compilation/constant-folding.ts:22–36

(n: number)

/**
 * The shader spelling of a NON-FINITE value (`NaN`, `±∞`).
 *
 * Neither GLSL nor WGSL has a `NaN` or an `Infinity` LITERAL — but both can
 * MAKE those values from a bit pattern, and the GPU target already routes the
 * masked (`When` / `Which` fall-through) NaN through exactly that mechanism
 * (`gpuNaN` in `gpu-target.ts`, which delegates here so there is exactly one
 * spelling). A non-finite CONSTANT is the same value reached by another route,
 * so it gets the same spelling instead of failing the compilation.
 *
 * GLSL goes through the overridable preamble helpers `_gpu_nan()` /
 * `_gpu_inf()` — one symbol a host can redefine without touching the generated
 * code. WGSL uses an inline `bitcast`, matching the existing WGSL NaN
 * convention (its spelling is pinned by `compile-wgsl.test.ts`).
 *
 * Infinity is a BIT PATTERN, never `1.0 / 0.0`: a fast-math driver is licensed
 * to fold a division by a constant zero (ANGLE→Metal fast-math already
 * destroys compensated arithmetic in this project), and a bit pattern is not
 * foldable.
 */